On 10/15/09 9:42 AM, "Steve Loughran" <[email protected]> wrote: > It's an interesting Q as to what is better, fewer nodes with more > storage/CPU or more, smaller nodes. > > Bigger servers > * more chance of running code near the data > * less data moved over the LAN at shuffle time > * RAM consumption can be more agile across tasks. > * increased chance of disk failure on a node; hadoop handles that very > badly right now (pre 0.20 -datanode goes offline) > > Smaller servers > * easier to place data redundantly across machines > * less RAM taken up by other people's jobs > * more nodes stay up when a disk fails (less important on 0.20 onwards) > * when a node goes down, less data to re-replicate across the other > machines > > 1. I would like to hear other people's opinions,
- Don't forget the about the more obvious things: if you go with more disks per server, that also means likely means less controllers doing IO. - Keep in mind that fewer CPUs/less RAM=less task slots available. While your workflow may not be CPU-bound in the traditional sense, if you are spawning 5000 maps, you're going to need quite a few slots to get your work done in a reasonable time. - To counter that, it seems we can run more tasks-per-node in LI's 2U config than Y!'s 1U config. But this might be an apples/oranges comparison (LI uses Solaris+ZFS, Y! uses Linux+ext3). > 2. The gridmix 2 benchmarking stuff tries to create synthetic benchmarks > from your real data runs. Try that, collect some data, then go to your > suppliers. +1
